MLMVN as an intelligent image filter
Igor N. Aizenberg, Alan Ordukhanov, Fionntan O'Boy · 2017
In this paper, we employ the multilayer neural network with multi-valued neurons (MLMVN) as an intelligent image filter. MLMVN is a powerful complex-valued feedforward neural network, which has shown its high efficiency in solving various classification and prediction problems. Here we employ it as an intelligent filter. MLMVN containing a single hidden layer is used to filter overlapping patches taken from a noisy image. Then the resulting image is obtained by averaging the intensities in all overlapping pixels. MLMVN is trained using a number of n × n patches from different images to obtain a clean patch from a noisy one. Then MLMVN is used to process overlapping patches taken from a noisy image. It is shown that this approach works better when an m × m patch is resulted from an n × n patch where m <; n, thus when the inner area of the patch is filtered. It is shown that this filter is robust, since it performs well on different images, which did not participate in the learning process. It terms of PSNR, this approach shows results better or comparable with BM3D filter commonly recognized as the best nonlinear filter. A specific advantage of the presented approach is its ability to preserve small details carefully. A batch learning algorithm (MLMVN-LLS) is employed to train the network.